Diversity and Serendipity in Recommendations with minimum information

For past two months I have been exploring in improving the diversity and serendipity of the recommendation systems. This actually came to me while I was working at eBay as a Research Intern. I realized I don’t qualify as a active user of the system. I don’t buy much from the site but when I buy I spend vast amount of time in the site to make that purchase. So I give out less information to the system just from the buying item perspectives. And I realized eBay gives me most obvious recommendations. So I started to explore how can I really leverage the diversity and serendipity of the recommendations that eBay gives me.

I have been reading and also have been exploring potential algorithm tweaking. We know Item-Item is fast but with Item-Item we loose much of the diversity and serendipity. But is there a solution? Is it possible to do a mix model?